Scientific Reform and Visual Data Science: Retiring the EDA/CDA dichotomy

Concerns around the replicability of published scientific findings has prompted much introspection into the way in which scientific knowledge is produced. To address issues of data fishing, searching exhaustively for discriminating patterns in a dataset, picking and then publishing those that are statistically significant, an argument is made that research findings should only be claimed through pre-registered confirmatory data analyses. Pre-registration studies are, though, somewhat inimical to the more informal research environments typical of modern applied data analysis (‘Data Science’). In this talk I enumerate some of these challenges and demonstrate, through an analysis of road crash data in the UK, how nascent visualization techniques can be used to navigate and inject statistical rigour into contemporary data analysis environments.